Looktap MCP

What the Looktap MCP MCP server does

Looktap MCP MCP server connects an AI agent to mobile apps running on iOS simulators or devices and Android emulators or devices. It exposes actions for reading the current screen, interacting with visible controls, collecting diagnostic output, and controlling application or device state.

Screen inspection returns a compact text listing of visible elements, including references, types, coordinates, dimensions, labels, and values. Agents can use those results to target controls by label or by a reference from the most recent screen read. Matching favors exact labels and rejects ambiguous matches rather than choosing arbitrarily.

How it works

The project is built around a command-line workflow in which each operation performs one action and prints a small amount of plain text. Actions can request a screen listing in the same call, and output can also be requested as JSON. There is no session setup before the first command, and the README describes no server process that users need to start.

Supported targets include native, React Native, and web content displayed in Safari, in-app browsers, or web views. The tool can work with real phones as well as simulated or emulated devices. Device commands can boot a named simulator or emulator, identify connected devices, and shut down a booted target.

Setup and configuration

Install requirements include Node.js 18 or later. The README states that simulators, emulators, and Android phones do not require additional project dependencies, while a real iPhone receives a small test runner that is built for it. Helper components start on first use.

The available material does not specify environment variables or provide a complete MCP client configuration example. It does document a Claude Code skill that can be installed with one command, but the exact installation command is not included here. The underlying package is published as @looktap/cli.

Tools and capabilities

The Looktap MCP MCP server supports capabilities in these areas:

• Read screen elements and values as text.
• Tap by label, element reference, or coordinates.
• Type into focused fields, replace field contents, send keys, and hide the keyboard.
• Swipe, scroll to labels, long-press, zoom, go back, return home, and open deep links.
• Launch, stop, install, reset, uninstall, and inspect applications.
• Capture PNG screenshots, screen-change bursts, and MP4 recordings.
• Read filtered application logs, crash reports, Android ANRs, and low-memory kills.
• Inspect CPU, memory, threads, frame rate, jank, and frame-time metrics.
• Simulate biometrics, location, appearance, text size, rotation, network offline mode, permissions, and selected notifications.
• Assert visible text or values with expect, run flow files as tests, and produce JUnit reports.

On Android, codegen can record a flow while the device is being used. Actions can wait for a label to appear or disappear instead of relying only on fixed sleeps.

Limitations and notes

Platform support differs by capability. For example, some simulated conditions are platform-specific, and the README identifies iOS crash reports and Android crashes, ANRs, and low-memory kills separately. The available material does not define a complete capability matrix for every device type.

The project reports a focus on compact agent output: screen reads are approximately 1 KB of text, while screenshots are saved to a file path. A scale option can reduce screenshot size. Locked devices and system prompts are reported rather than treated as ordinary app content, and stale element references or ambiguous labels produce errors.

Looktap MCP MCP server is licensed under AGPL-3.0. The README names Claude Code, Cursor, and Codex as suitable AI coding-agent use cases; it does not document support details for every MCP client.